High-dimensional macroeconomic stress testing of corporate recovery rate

We investigate macroeconomic stress testing frameworks for corporate bond recovery rate analysis using machine learning techniques. In doing so, we simulate the macroeconomic effect of a broad range of 182 macroeconomic variables extracting key factors with methods such as (sparse) principal component analysis and sparse group least absolute selection and shrinkage operation (LASSO). Using the adverse stress testing scenario from the US Federal Reserve as the benchmark, we demonstrate that our least squares-support vector regression model produces sensible and potentially valuable risk measures such as value-at-risk and conditional value-at-risk for recovery rates during periods of macroeconomic stress.